FEAT: Chain-of-Thought Rendering for Attack Results and Scenario Results

doc/code/output/0_output.ipynb

@@ -515,6 +515,39 @@
    "cell_type": "markdown",
    "id": "13",
    "metadata": {},
+   "source": [
+    "## Including Reasoning Summaries\n",
+    "\n",
+    "Reasoning-summary output is opt in: the conversation, attack-result, and scenario-result\n",
+    "helpers hide reasoning by default. For OpenAI Responses targets, PyRIT renders\n",
+    "provider-generated reasoning summaries exposed by OpenAI, not raw hidden chain-of-thought.\n",
+    "\n",
+    "- Pretty output shows the `<reasoning-summary>` and `</reasoning-summary>` tags in subdued gray.\n",
+    "- Attack Markdown output shows visible escaped `\\<reasoning-summary\\>` and\n",
+    "  `\\</reasoning-summary\\>` tags.\n",
+    "\n",
+    "```python\n",
+    "from pyrit.output import output_attack_async, output_conversation_async, output_scenario_async\n",
+    "\n",
+    "# Direct conversation\n",
+    "await output_conversation_async(messages=conversation, include_reasoning_trace=True)\n",
+    "\n",
+    "# Attack result (Pretty or Markdown)\n",
+    "await output_attack_async(attack_result, include_reasoning_trace=True)\n",
+    "await output_attack_async(attack_result, format=\"markdown\", include_reasoning_trace=True)\n",
+    "\n",
+    "# Existing scenario result (Pretty only)\n",
+    "await output_scenario_async(scenario_result, include_reasoning_trace=True)\n",
+    "```\n",
+    "\n",
+    "For a scenario, enabling reasoning expands every contained attack result so its summaries\n",
+    "can be rendered. This can produce verbose output for large scenario runs."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "14",
+   "metadata": {},
    "source": [
     "## Printing Scores\n",
     "\n",
@@ -524,7 +557,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "14",
+   "id": "15",
    "metadata": {},
    "outputs": [
     {
@@ -551,7 +584,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "15",
+   "id": "16",
    "metadata": {},
    "source": [
     "## Sinks — Redirecting Output\n",
@@ -565,7 +598,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "16",
+   "id": "17",
    "metadata": {},
    "outputs": [
     {
@@ -604,7 +637,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "17",
+   "id": "18",
    "metadata": {},
    "source": [
     "### Available Sinks\n",
@@ -621,7 +654,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "18",
+   "id": "19",
    "metadata": {},
    "source": [
     "## Using Printers Directly\n",
@@ -634,7 +667,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "19",
+   "id": "20",
    "metadata": {},
    "outputs": [
     {
@@ -675,7 +708,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "20",
+   "id": "21",
    "metadata": {},
    "source": [
     "### `render_async` vs `write_async`\n",
@@ -689,7 +722,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "21",
+   "id": "22",
    "metadata": {},
    "outputs": [
     {
@@ -728,7 +761,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "22",
+   "id": "23",
    "metadata": {},
    "source": [
     "## Architecture Overview\n",
@@ -785,7 +818,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "23",
+   "id": "24",
    "metadata": {},
    "source": [
     "## Convenience Functions Reference\n",
@@ -805,7 +838,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "24",
+   "id": "25",
    "metadata": {},
    "source": [
     "## Extending the Printer Module\n",

doc/code/output/0_output.py

@@ -143,6 +143,34 @@
 # print the conversation using the print conversation helper
 await output_conversation_async(messages=conversation)  # type: ignore
 
+# %% [markdown]
+# ## Including Reasoning Summaries
+#
+# Reasoning-summary output is opt in: the conversation, attack-result, and scenario-result
+# helpers hide reasoning by default. For OpenAI Responses targets, PyRIT renders
+# provider-generated reasoning summaries exposed by OpenAI, not raw hidden chain-of-thought.
+#
+# - Pretty output shows the `<reasoning-summary>` and `</reasoning-summary>` tags in subdued gray.
+# - Attack Markdown output shows visible escaped `\<reasoning-summary\>` and
+#   `\</reasoning-summary\>` tags.
+#
+# ```python
+# from pyrit.output import output_attack_async, output_conversation_async, output_scenario_async
+#
+# # Direct conversation
+# await output_conversation_async(messages=conversation, include_reasoning_trace=True)
+#
+# # Attack result (Pretty or Markdown)
+# await output_attack_async(attack_result, include_reasoning_trace=True)
+# await output_attack_async(attack_result, format="markdown", include_reasoning_trace=True)
+#
+# # Existing scenario result (Pretty only)
+# await output_scenario_async(scenario_result, include_reasoning_trace=True)
+# ```
+#
+# For a scenario, enabling reasoning expands every contained attack result so its summaries
+# can be rendered. This can produce verbose output for large scenario runs.
+
 # %% [markdown]
 # ## Printing Scores
 #

pyrit/output/attack_result/base.py

@@ -3,7 +3,7 @@
 
 from abc import abstractmethod
 
-from pyrit.models import AttackOutcome, Message, Score
+from pyrit.models import AttackOutcome, AttackResult, Message, Score
 from pyrit.output.base import PrinterBase
 
 
@@ -18,6 +18,31 @@ class AttackResultPrinterBase(PrinterBase):
     Thin-client implementations can fetch data via REST endpoints.
     """
 
+    @abstractmethod
+    async def render_async(
+        self,
+        result: AttackResult,
+        *,
+        include_auxiliary_scores: bool = False,
+        include_pruned_conversations: bool = False,
+        include_adversarial_conversation: bool = False,
+        include_reasoning_trace: bool = False,
+    ) -> str:
+        """
+        Render an attack result.
+
+        Args:
+            result (AttackResult): The attack result to render.
+            include_auxiliary_scores (bool): Whether to include auxiliary scores. Defaults to False.
+            include_pruned_conversations (bool): Whether to include pruned conversations. Defaults to False.
+            include_adversarial_conversation (bool): Whether to include the adversarial conversation.
+                Defaults to False.
+            include_reasoning_trace (bool): Whether to include reasoning traces. Defaults to False.
+
+        Returns:
+            str: The rendered attack result.
+        """
+
     @abstractmethod
     async def _get_conversation_async(self, conversation_id: str) -> list[Message]:
         """